this programme 5 data values n 1 , n 2 , x1, x 2 , and the required CI need to be entered. For
this example the 5 data values are:
data a; ** Enter 5 data values on following 5
lines **;
n1=110; ** n1 is sample size for sample
1 **;
n2=108; ** n2 is sample size for sample
2 **;
x1=96; ** x1 is relevant count for sample
1 **
x2=72; ** x2 is relevant count for sample
2 **;
CI=95; ** Required confidence interval e.g. 95,
90 **;
Output from the procedure PROC PRINT, which simply prints the values of the variables
is
OBS N1 N2 X1 X2 CI P1 P2 ALPHA SEDIFF Y
1 110 108 96 72 95 0.87273 0.66667 0.05 0.055384 1.95996
LOWERCI UPPERCI
0.10 0.31
From this output the computed 95 per cent CI, 0.10 to 0.31, is the same as the interval
values calculated in the preceding section. Computer output for a test of the second null
hypothesis, that of no difference in the proportion of respondents between Q1 and Q3
who endorse the category very much/much, is shown below:
OBS N1 N2 X1 12 P1 P2^ CI ALPHA Y
1 110 109 96 71 0.87273 0.65138 95 0.05 0.055616 1.95996
LOWERCI UPPERCI^
0.11 0.33
Interpretation of Computer Output
Interpretation is straightforward, in neither comparison does the 95 per cent confidence
interval include the value zero. We can therefore conclude with 95 per cent certainty that
the population proportions of affirmative responses to the two questions, Q1 and Q2 and
to the questions Q1 and Q3 (two comparisons) are significantly different, p<0.05. The
estimate of the difference in the percentage of respondents who endorse the statement that
they ‘very much or much’ believe that the educational psychologist should be involved in
providing advice on materials compared with those who endorse the statement ‘should be
involved with in-service work’ is 20.6 per cent (0.873−0.667×100). The 95 per cent
confidence interval ranges from 10 per cent to 31 per cent, and the standard error of the
difference is 5.5 per cent.
It is good practice to note both the confidence interval and the p value (and any test
statistic values) when reporting results.
Statistical analysis for education and psychology researchers 188